Pye, C. 1993. A Cross-Linguistic Approach to the Causative Alternation, in Yonata Levy (Ed.), OtherChildren, Other Languages, pp. 243-263. Hillsdale, NJ: Erlbaum.
A Crosslinguistic Approach to the Causative Alternation
Clifton Pye
Linguistics Department
University of Kansas
Lawrence, KS 66045
A Crosslinguistic Approach to the Causative Alternation
One of the bedrocks of current linguistic investigation is
that there is a fairly direct mapping between thought and
language. The assumption is that at some level of the grammar,
whether it is d-structure (Chomsky 1981), argument structure
(Grimshaw 1990) or lexical conceptual structure (Hale & Keyser
1986), there is a tie between language structure and a more
universal cognitive perception of events. Perlmutter & Postal's
(1984) Universal Alignment Hypothesis (UAH) and Baker's (1988)
Uniformity of Theta Assignment Hypothesis (UTAH) are concrete
expressions of this assumption. These hypotheses, with Chomsky's
(1981) Projection Principle (or it's extended variants) impose
meaningful constraints on verb argument structures. A universal
or uniform association of thematic roles to syntactic positions
would eliminate crosslinguistic variation in verb argument
structures.
Ideally, children would be able to use such principles in
constructing their initial lexical entries for verbs and so avoid
confusion from hearing well-formed sentences with null arguments
or ill-formed sentences with missing arguments. On hearing a
sentence such as 'The stick broke,' the UAH or UTAH would tell a
child that the NP 'the stick' was in the direct object position
at some level of the syntactic derivation since it bears the
thematic role of theme. The linguistic principles in combination
with the primary linguistic data would provide the child with all
the information necessary to construct a syntactic derivation for
the sentence.
Things are not quite this simple. Some recent work has
explored the implications of Perlmutter's (1978) Unaccusative
Hypothesis for the UAH and UTAH. The Unaccusative Hypothesis
claims that it is necessary to distinguish between intransitive
verbs such as run, jump and arrive that are unergative and
intransitive verbs such as melt, slide and roll that are
unaccusative. A host of tests in many languages suggest that
unergative and unaccusative verbs have different initial argument
alignments. If the UAH and UTAH are correct then all languages
should classify the same verbs as unergative or unaccusative.
Since Carol Rosen's initial survey (1984) linguists have debated
the degree of crosslinguistic uniformity among intransitive verbs
(Grimshaw 1990, Levin & Rappaport Hovav 1992, Perlmutter 1989?,
Van Valin 1990). Certain subclasses of verbs (e.g. psych verbs
and verbs for bodily processes) exhibit more crosslinguistic
variation in argument structure than other subclasses (e.g.
motion verbs), but a final assessment of the degree of cross-
linguistic variation is not available.
Another problem is that languages commonly employ one or
more processes that alter the argument structure of verbs.
Examples of such processes include the passive, antipassive,
causative, dative, locative, conative, middle and applicative
alternations (c.f. Baker 1988). If such processes worked in a
uniform manner they would not create a problem since a child
would have direct evidence on how each process affected verb
argument structure. Passives, for example, eliminate the direct
expression of agent roles at the level of syntax, while
antipassives eliminate the direct expression of theme roles at
the syntactic level. The problem is that there is a great deal
of crosslinguistic variation in the productivity of each process.
Every language contains verbs that do not undergo a
particular process. The verbs have, resemble and want cannot be
passivized in English, for example. Children could solve this
problem if the lexical exceptions to a process involved a small
set of thematic roles or semantic features (Green 1974).
Pinker's (1989) criterion-based hypothesis applies such a
procedure to account for the acquisition of the passive,
causative, locative and dative alternations in English. Such
solutions, though, call for crosslinguistic study of argument
structure alternations. The lexical exceptions to argument
changing processes should fall within discrete subclasses of
verbs across languages. These subclasses, in turn, would provide
useful data for determining the relevant restrictions on the
operation of the argument changing processes.
The Causative Alternation
In this paper I will show that detailed crosslinguistic
data provides an insight into the manner in which the structure
of the verb lexicon affects the acquisition of transitivity
alternations. I will pay particular attention to the causative
alternation. Many verbs alternate between intransitive and
transitive syntactic forms to indicate the cause of a patient
argument's change of state. The mystery is that not all
intransitive verbs alternate in the same way. Bowerman (1974)
pointed out that children learning English sometimes extend the
causative alternation to verbs that alternate in other ways.
Thus, Bowerman reports such examples as Christy (2;9) saying 'I
come it closer so it won't fall' instead of 'I made it come
closer' or 'I moved it closer.' My own son has produced several
of these errors including 'Mommy has to talk the king' for 'Mommy
has to make the king talk' said when he was 3 years and 7 months
old.
Pinker (1989) underlines the dilemma such constructions
raise for language acquisition theories. Once children determine
that a particular transitivity alternation is productive they may
extend the alternation to new verbs. Such extensions are
unacceptable for English verbs like 'come', but there is no
record of parent tutorials correcting children who produce these
forms. There is no other obvious means that children could rely
upon to 'cut back' unacceptable extensions. Thus, there is no
logical procedure children can use to acquire the causative
alternation.
Pinker outlines three possible sources of children's errors.
They could be applying a lexical rule too broadly, and fail to
notice narrow-based semantic constraints that restrict the
lexical rule. Alternatively, children may retrieve the wrong
verb stem under pressure from the discourse. A third possibility
is that children have not yet acquired an adult semantic
representation for some verbs and thus misuse these verbs.
Pinker feels that children stop producing causative errors when
they learn the meaning of each verb, when they are better at
retrieving verbs, and when they learn the proper restrictions on
the causative alternation.
In this paper I report the results of a pilot elicitation
study that I conducted with K'iche' Maya children from the
Western Highland region of Guatemala. The K'iche' children
showed a different pattern of causative alternations than their
English-speaking peers which indicates that language-specific
structures influence the acquisition of the causative alternation
more than the semantic structure of individual verbs. The
following section of the paper sets out the system of causative
alternations found in the K'iche' language. The next section
presents acquisition data on the K'iche' causative from a
longitudinal study of children. I then present the results of an
elicitation study on the causative alternation and discuss its
findings. In the last section of the paper I discuss the
implications of this study for different theoretical proposals
concerning the acquisition of the causative alternation.
The Causative Alternation in K'iche' Maya
K'iche' is a Mayan language spoken by a million inhabitants
of the Western Highlands of Guatemala. K'iche' has an
agglutinating morphology which reflects the distinction between
transitive and intransitive verbs in several respects (see 1).
The language has an ergative cross-referencing system on the
verb, so intransitive verb subjects and transitive verb objects
are marked with an absolutive marker while transitive verb
subjects receive a distinctive ergative marker. All verbs also
require a special clause-final termination which distinguishes
between transitive and intransitive verb stems (Pye 1983).1
Intransitive verbs in simple, declarative sentences have the
clause-final termination /-ik/ while 'root' transitive verbs in
the same sentences have the termination /-oh/ or /-uh/ and
'derived' transitive verbs have the termination /-Vj/ where V may
be either /i, e, a, o, u/.
(1) Transitive verbs Intransitive verbs
a. k-at-inw-il-oh c. k-at-b'e:-ik
INCOMP-2A-1E-see-TTV INCOMP-2A-go-IV
'I see you.' 'You are going.'
b. k-0-a-kuwi:-j d. k-0-taq'en-ik
INCOMP-3A-2E-hurry-TV INCOMP-3A-PROGRESSIVE-IV
'You are hurrying.' 'It is-ing2.'
K'iche' uses the suffix /-isa/ to derive the causative form of
intransitive verbs. Examples of this causative construction are
shown in (2). In K'iche' the causative suffix can only be added
to intransitive verb stems, unlike Berber, Japanese and Korean
where it is also possible to add a causative affix to transitive
verb stems.
(2) K'iche' causative verbs with /-isa/
Intransitive Form Causative Form
a. k-0-poqow-ik k-0-a-poqow-isa:-j
INCOMP-3A-boil-IV INCOMP-3A-2E-boil-CAUSE-TV
'It is boiling.' 'You are boiling it.' ( = cause to boil)
b. q'alaj k-0-in-q'alaj-isa:-j
clear INCOMP-3A-1E-clear-CAUSE-TV
'It is clear.' 'I will clarify things.' ( = cause to be clear)
Although the causative construction is very productive in
K'iche' it is not completely so. There are two classes of
intransitive verbs in K'iche' which do not take the causative
affix. The first of these exceptional classes uses another means
of deriving a transitive verb stem. I will refer to this class
of verbs collectively as the 'zero class' although the examples
in (3) show that this set of verbs uses several different
derivational processes. The transitive verb in (3a) has a
polysyllabic stem and so takes a 'derived' transitive termination
marker /-j/3. Its intransitive counterpart deletes the final
vowel from the stem and changes the final consonant before adding
the intransitive termination. The verb pair in (3b) simply
switch termination markers without changing any other part of the
verb stem, while the intransitive verb in (3c) adds the
intransitive termination to the whole transitive verb, including
its derived transitive termination as part of the intransitive
stem. I group these verbs together because they share the
feature of alternating between intransitive and transitive verb
forms by a derivational process that is distinct from the regular
affixal causative process (c.f. Dayley 1985).
(3) K'iche' zero class verbs
Transitive verbs Intransitive verbs
a. k-0-in-qasa:-j k-in-qaj-ik
INCOMP-3A-1E-go_down-TV INCOMP-1A-go_down-IV
'I am taking it down.' 'I am going down.'
b. x-0-in-tzaq-oh x-in-tzaq-ik
COMP-3A-1E-drop-TTV COMP-1A-fall-IV
'I dropped/lost it.' 'I fell.'
c. x-0-a-tzali:-j x-at-tzalij-ik
COMP-3A-2E-return-TV COMP-2A-return-IV
'You returned it.' 'You returned.'
I will dub the remaining set of exceptional intransitive
verbs the periphrastic class. This set of verbs does not permit
any derivational process to produce a simple transitive verb
stem. The only way to express a transitive notion with the
members of this set is to use a complex sentence which contains a
verb like -b'an 'do/make'. Examples of periphrastic verbs are
shown in (4).
(4) K'iche' periphrastic verbs
Intransitive verbs Periphrastic constructions
a. k-in-pet-ik k-0-in-b'an k-at-pet-ik
INCOMP-1A-come-IV INCOMP-3A-1E-do INCOMP-2A-come-IV
'I am coming.' 'I will make you come.'
b. k-in-muxan-ik k-0-in-b'an k-at-muxan-ik
INCOMP-1A-swim-IV INCOMP-3A-1E-do INCOMP-2A-swim-IV
'I am swimming.' 'I will make you swim.'
The K'iche' causative alternation is further complicated by
one additional factor. Transitive verbs may lose the direct
object with the addition of an absolutive antipassive affix.
K'iche' speakers use the absolutive antipassive to emphasize an
action rather than a result (Mondloch 1981). With many verbs,
the antipassive acts like an anticausative. Two examples
of the absolutive antipassive alternation are shown in (5).
(5) The K'iche' absolutive antipassive
a. k-0-u-chaku:-j k-0-chaku-n-ik
INCOMP-3A-3E-work-TV INCOMP-3A-work-ABS-IV
'He/she is working it.' 'He/she is working.'
b. x-in-a-tze'-j x-at-tze'-n-ik
COMP-1A-2E-laugh-TV COMP-2A-laugh-ABS-IV
'You made me laugh4.' 'You laughed.'
To put it mildly, the combination of productive causative and
antipassive alternations plus a good number of lexical exceptions
should create considerable problems for any child so unfortunate
as to be faced with the prospect of learning K'iche'.5 Such
complexity provides the perfect testing ground for competing
hypothesizes about the sources of children's derivational rules.
If children are conservative learners then they should be
unwilling to produce a causative alternation until they have
learned the proper lexical form. In this case their language
productions should be error-free (Baker 1979). Another
possibility would be that children adopt a simple lexical schema
as a ready-made means of producing the causative alternation
(Braine et al. 1990). Such a schema would result in a multitude
of overgeneralization errors for K'iche' such as using simple
intransitive verb stems in transitive sentences. Finally,
children might rely upon some type of semantic classification
when applying the causative alternation (Green 1974; Pinker 1989;
Slobin 1985). In this case K'iche' children should initially
behave like children learning English and have difficulty with
similar ranges of semantic subclasses of verbs.
Analysis of K'iche' Language Samples
Spontaneous language samples suggest that the causative
derivation is a relatively late acquisition for K'iche' children.
I have found that by 2;10 the children are beginning to produce
examples of causativized verbs (Pye 1992). Children learning
English begin to produce examples of causativized verbs around
2;2, while children learning Turkish apparently begin producing
causativized verbs around 2;3 (Aksu-Koç & Slobin 1985). The
K'iche' children's causativized verbs alternate with the
intransitive verb forms, sometimes in the same session. Their
alternations are evidence that the children apply a productive
morphological alternation rather than using the same verb form in
transitive and intransitive contexts. However, the children do
not use the causative affix very frequently in the spontaneous
data. My three primary subjects produced a total of 24 affixal
causativized verb tokens in 20,103 utterances. The children also
used intransitive and transitive verbs from both the zero and
periphrastic verb classes without adding the causative affix to
them. They did not add the causative affix to any transitive
verb stems.
I have found two examples in which one of the children
failed to produce the causative affix. The first occurred when
Al Cha:y was 2 years and 10 months old and had a mean utterance
length of 1.92 morphemes. Her older sister told her to tell me
to turn on my tape recorder using a verb which requires the
causative affix in this context. Al Cha:y produced this
utterance as k'at e laya (= chak'atisaj le: aradio). Al Cha:y's
production lacks the causative affix and literally means 'Your
radio is burning.' The causativized form of this verb undergoes
a semantic shift from 'burn' to 'turn on'. Two weeks later my
assistant and I were picking the girls up and swinging them
around. We had just done this with Al Cha:y's sisters and Al
Cha:y wanted to be next. In this context she produced the
utterance lij in (= kinawalijisaj in). If she wanted to be
picked up she should have used the verb with the causative affix
(walij-isa-j). Her utterance literally means 'I am going up'
which is a possible, though less likely interpretation in this
context. Later in the same tape Al Cha:y succeeded in producing
a verb with the causative affix (paq'ixaj chik = kinapaqilisaj
chik 'Pick me up again').
This result agrees to some extent with acquisition data from
other languages. Bowerman (1974) has provided many examples of
such causative overgeneralizations in English, and Berman (1982)
noted examples from the initial stage of learning Hebrew. The
number of examples these authors cite lead one to believe that
causative overgeneralizations are fairly common in the initial
stage of language acquisition. Unfortunately it is impossible to
calculate base rates of overgeneralization on the basis of the
data provided. My finding of 2 overgeneralizations in 20,103
utterances is comparable to Maratsos' (1979) estimation that
Bowerman's 100 or so examples were culled from approximately
750,000 utterances. However, it would be more accurate to use
the number of transitive contexts in which affixal causative
verbs appear rather than the total number of utterances to
calculate overgeneralization rates. For the K'iche' samples, the
finding of 2 overgeneralizations in 26 contexts increases the
rate of overgeneralization to 8 percent. This rate suggests that
K'iche' children have an accurate knowledge of verb transitivity
that is susceptible to occasional lapses in performance (c.f. Pye
1987). The verb argument structures in their verbal lexicon are
essentially correct.
An Experimental Study
Experimental studies provide the best means of estimating
overgeneralization rates and thereby children's knowledge of the
causative alternation. Such studies provide a convenient way to
collect information on a large number of causative verb forms in
a short amount time. An experimental study of the K'iche'
causative alternation provides the best way of collecting
information on children's knowledge of all three verb categories
as well as a way to determine whether children make the same
types of overgeneralizations with the verbs in each category.
One drawback to such a study might be that it would induce
children to make more than the usual number of
overgeneralizations. A control for this problem is to test older
subjects. If the experiment induces adult subjects to
overgeneralize then it is not looking at lexical competence. If
there is a gap between the children's rate of overgeneralization
and that of the adults then one can estimate the degree of
priming the experiment is creating and eliminate that from the
estimation of the children's overgeneralization rates. I put
together a pilot test of the K'iche' causative alternation using
verbs from all three groups. The verbs are shown in (6).
(6) K'iche' verbs from causative elicitation study
Morphological Zero-derivation verbs Periphrastic verbs
xojow-isa dance-CAUSE qaj-ik go_down-IV muxan-ik swim-IV
aq'an-isa climb-CAUSE sutin-ik turn-IV wakat-ik walk-IV
ch'aqt-isa wet-CAUSE el-ik leave-IV pet-ik come-IV
noj-isa full-CAUSE wul-ik destroy-IV
atin-isa bathe-CAUSE
Method
Our initial pilot testing had shown that we could induce
children to produce more causative forms if we began with some
familiar causativized verbs. Therefore, we began the test by
eliciting the causative forms for the verbs xojow 'dance' and
aq'an 'climb'. Thereafter we alternated between the different
classes of verbs. We used the same order for each child (xojow
'dance', aq'an 'climb', ch'aqt 'wet', qaj 'do down', sutij
'turn', wul 'destroy', pet 'come' noj 'full', atin 'bathe', el
'leave', muxan 'swim' and wakat 'walk'). We used a set of
plastic farm animals as our stimulus items, primarily a mother
pig and two baby pigs. For example, our protocol for the verb
xojow 'dance' went:
'Kaxojow ri: aq i'. Kawiloh? Kaxojowik. Ma kaxojow taj
le: jun aq chik. Kara:j na luna:n kaxojow le: ra:l y ku'an
le: ri ri'. Jas ku'an le: nan che le: ra:l?'
'This baby pig is dancing. See? It's dancing. The other
baby pig is not dancing. Its mother wants her baby to dance
so she goes like this. What is she doing to her baby?'
If a child failed to respond we would repeat the action and again
ask what the mother was doing to her baby. If a child responded
that the baby was dancing, we would draw their attention to the
mother's action and again ask what the mother was doing to her
baby. If the child still could not say what the mother was
doing, we would record the response as a refusal and go on to the
next item. I was surprised to find that we had very little
difficulty eliciting transitive verbs from even our youngest
subjects in this manner. While one of us manipulated the animals
and delivered the monologue, the other would transcribe the
children's responses. In addition, all sessions were audio-
recorded.
Results
Table 1 shows what an ideal response pattern would look
like. Children should use the morphological form of the
causative with morphological causative verbs, the zero form with
zero derivational verbs, and the periphrastic form with
periphrastic verbs. Anything else would count as an
overgeneralization.
Table 1. Ideal Response Pattern
Type Used
Type Morphological Zero Periphrastic
Morphological X
Zero X
Periphrastic X
We elicited quite a range of responses from our subjects.
Besides the expected (adult) responses, the children used other
transitive verbs, other causativized verbs, periphrastic
responses, the intransitive verb form or another intransitive
verb. Their responses are shown in Table 2. I calculated the
percentage of overgeneralizations based on the types of errors
for each class of verbs. For the morphological causative, only
the use of a zero form would count as an overgeneralization. For
the zero verb class, only the use of a causative affix /-isa/
would count as an overgeneralization, and for the periphrastic
verbs, the use of either the causative affix or a zero derivation
would count as an overgeneralization.
Table 2. Causative Data
4,5,6,7 years 8,9,10,11 years 12,13 years (N = 11) (N = 62) (N = 7) Morph Zero Other Morph Zero Other Morph Zero Other
Morphological
xojowisa:j 11 58 4 7atinisa:j 11 55 7 6 1aq'anisa:j 11 42 20 4 3nojisa:j 8 3 36 26 6 1ch'aqtisa:j 1 10 32 30 3 4 Percent 0% (0/42) 0% (0/223) 0% (0/26)
Zero qasa:j 3 4 4 12 31 19 1 4 2suti:j 11 4 55 3 1 6esa:j 8 3 46 16 2 5wuli:j 3 8 16 35 1 6
Percent 10% (3/29) 10% (16/164) 13% (2/15)
Periphrastic muxanik 1 7 3 1 28 33 1 1 5wakatik 1 10 3 1 58 1 6petik 3 23 5
Percent 35% (9/25) 24% (33/139) 16% (3/19)
Total Percent 10% (11/107) 8% (49/575) 8% (5/63)
Discussion
Our most significant finding is that we actually succeeded
in eliciting some causative overgeneralizations from the
children. We have had some difficulty eliciting passive
sentences from K'iche' children in previous studies (Pye &
Quixtan Poz 1989) so I was relieved to find that it was fairly
easy to get the children to talk about causative actions. Some
children added the causative affix to the zero class verbs qajik
'go down' and sutinik 'turn' as well as the periphrastic class
verbs muxanik 'swim' and wakatik 'walk'. It was also a surprise
to see that the children applied the zero derivation to verbs in
the periphrastic class as well as using the 9 regular causative
derivation. For muxanik 'swim' their favorite zero derivation
was muxa:j, while their zero derivation for wakatik 'walk' was
wakati:j.
I was not prepared to find the children overgeneralizing
these verbs so frequently. I had expected the children to
overgeneralize in 8 to 10 percent of their responses based on my
previous estimation from longitudinal samples and reports in the
literature (Cazden 1968). Experimental studies such as this one
seem to elicit higher rates of overgeneralization. Maratsos et
al. (1987) report a mean overgeneralization rate of 26% while a
recent study by Braine et al. (1990) found their subjects
overgeneralized the English causative alternation to intransitive
verbs in 39% of the trials. Seventy-three percent (8/11) of the
youngest K'iche' subjects extended the causative to the verb
muxanik 'swim', while 43% (3/7) extended the morphological
causative to the verb qajik 'go down'. Only 33% (2/6) of the
oldest subjects extended the causative to the verb muxanik
'swim'.
There is no indication of an order effect in the children's
responses. The final verbs on the test muxanik 'swim' and
wakatik 'walk' had a large number of overgeneralizations, but the
verb that immediately preceded them (elik 'leave') was not
overgeneralized by any of the children. The children produced
overgeneralized forms of two of the verbs in the zero class
(qajik 'go down' and sutinik 'turn') which were the fourth and
fifth items on the test. However, none of the children
overgeneralized the sixth and seventh items on the test (the
verbs wulik 'destroy' and petik 'come'). These trends suggest
that the children's responses reflected the state of their
knowledge of the individual lexical items rather than a simple
priming effect from the order of elicitation.
I was especially surprised that we succeeded in eliciting
causative overgeneralizations from 13-year-olds. An assumption
has crept into the literature that all the interesting
developments in syntax occur before 5;0. Pinker (1989.289)
states that Christy made such overgeneralizations over a period
of six years, from 2;1 to 7;11. Braine et al. only tested 2- and
4-year-olds in addition to adults. The K'iche' data shows that
the acquisition of lexical alternations is not completed in all
languages by 8;0. This, of course, raises the learnability issue
of exactly what mechanism would operate over such an extended
period of time. The slow rate of progress rules out a
maturational or grammatical change since such changes would lead
to more abrupt 'across-the-board' restrictions on the causative.
There was a striking difference in the children's
willingness to produce transitive versions of individual verbs.
The children were quite happy to supply causativized versions of
the verbs xojow 'dance', aq'an 'climb', and atin 'bathe', but had
real trouble finding a way to causativize ch'aq 'wet' and to a
lesser extent noj 'full'. This result shows that our experiment
was not equally successful in eliciting the causative forms of
all the verbs. Subjects often responded to probes on ch'aq 'wet'
with the causative form of atin 'bathe' since bathing someone is
more typical than wetting them.
Even more striking was the difference in the children's
overgeneralizations of the verbs in each class. Three of the
youngest subjects overgeneralized the verb qajik 'go down' while
8 of them overgeneralized the verb muxanik 'swim'. None of them
overgeneralized the verbs sutinik 'turn', elik 'leave' and petik
'come'. In fact we stopped using the verb petik 'come' in our
experiment because the K'iche' children were unwilling to
causativize it and it seemed to lead to more frustration on their
part when we kept probing for it.
This data suggests another important difference between
children learning K'iche' and those learning English. Pinker
(303) reanalyzed Bowerman's (1982) data and found that children
learning English most frequently causativized the verbs come, go,
fall, rise, and drop. The K'iche' children did their best to
avoid causativizing the verb petik 'come' while most frequently
overextending the causative derivation to the verbs qasaj 'go
down', muxanik 'swim', sutinik 'turn' and wakatik 'stroll'.
Berman (1982) reports that Hebrew-speaking children
overgeneralize the intransitive forms of the verbs see, eat,
move, sit, hurt, go down, get up (179) and Aksu-Koç & Slobin
report errors with the Turkish verbs burn and get up (1985.849).
The data suggests significant crosslinguistic differences in the
verbs children overgeneralize.
One possible explanation for these crosslinguistic
differences would be a difference in input frequencies. If
frequencies of the verbs differed significantly in the input
languages they might underlie differences in the children's
knowledge of each verb and therefore differences in the
overgeneralization rates for each verb. Children should find
more evidence for a verb's argument structure if that verb is
frequently produced by parents. Children might produce more
causative overgeneralizations with verbs that are less frequent
in the input.
I examined this issue by counting the different verbs in the
speech of the mothers in my spontaneous samples of K'iche'. I
had previously selected samples in which the mothers did a lot of
the talking to estimate the frequency with which the mothers used
various morphemes. I counted the number of times the mothers
used each of the verbs from the causative experiment. For a
comparison with English I counted the number of times Adam's
mother and Eve's mother used the English equivalents of the
K'iche' verbs in their first samples (Brown 1973).6 I used the
PAL computer program to extract the verbs from the mothers'
language samples (Pye 1987). The results are shown in Table 3.
Table 3. Token frequency of experimental verbs in speech to children
K'iche' Mothers English Mothers Verbs Al Tiya:n Al Cha:y A Carlos Adam Eve
xojow 'dance' 3aq'an 'climb' 2 1ch'aq 'wet' 2qaj 'go down' 1sutij 'turn' 2 6wulin 'destroy'pet 'come' 23 32 5 3 16noj 'fill'atin 'wash' 1 1el 'leave' 3 5 2 2 4muxan 'swim' 1wakat 'walk' 3
Total verb types 58 88 36 82 74Total verb tokens 285 417 290 523 789
I included tokens of both transitive and intransitive uses
of the verbs in my count. Thus, I counted uses of both pet
'come' and k'am 'bring' for the K'iche' mothers and their
equivalents for Adam's and Eve's mothers. I also excluded uses
of the English verbs that fell outside the range of meaning of
the K'iche' verbs. Adam's and Eve's mothers tended to use the
verb leave more frequently with a meaning of 'let it stay' rather
than 'go out of some place'. I only included the latter use in
my count. These samples may not be large enough to provide
incontrovertible evidence of a difference in input frequency
between K'iche' and English, but they are robust enough to rule
out a simple correlation between input frequency and causative
overgeneralizations.
There is a fair degree of consistency between the K'iche'
mothers which derives from their frequent use of the verbs pet
'come' and el 'leave' and their transitive counterparts k'am
'bring' and esa:j 'take out'. It is also true that the K'iche'
subjects in my experiment did not overgeneralize the causative
with these verbs. However, the K'iche' children did not
overgeneralize the causative with the verb wulin 'destroy'
either, despite its infrequency in the input. Adam's and Eve's
mothers both used the verb come the most frequently. This
finding is at odds with English-speaking children's frequent
overgeneralization of the causative alternation with come. It
seems as though differences in input frequencies will not account
for the crosslinguistic differences in children's willingness to
overgeneralize the causative with certain verbs although this
issue should be pursued farther with larger samples of parental
speech to children.
Another interesting finding was that the children did not
overgeneralize the intransitive verb forms to transitive
contexts. The classic observation from Bowerman is that children
use intransitive forms in transitive contexts e.g. Christy (2;9)
'I'm gonna just fall this on her.' In fact, this is the
phenomenon that Braine and Maratsos succeeded in eliciting from
their subjects. Berman (1982.179) reports that this is a
relatively frequent error for Hebrew-speaking children and Aksu-
Koç & Slobin (1985.849) provide an example of this type of error
in Turkish. Even though I found two examples of such an error in
my spontaneous K'iche' data we did not elicit a single example of
this sort from our K'iche' subjects. We did elicit a few
intransitive verbs from the children, but in these cases it is
clear that the children were using the verbs as intransitives.
The verbs have an intransitive morphology, and more telling, the
children only used these verbs with one argument.
One seven year old boy, for example, produced the sentence
in (7a). His sentence only contains one argument ('the little
pig') and the verb uses the regular third person absolutive
subject marker and the intransitive verb termination /-ik/. For
the trial with pet 'come' this same boy produced the sentence in
(7b). In this example the boy retains a transitive relation by
using the transitive verb yo' 'scold', and manages to convey the
right motion by adding an extra clause with the intransitive verb
pet 'come'. He displays his control of the verb morphology in
switching between an ergative subject marker on the transitive
verb yo' 'scold' and an absolutive subject marker on the verb pet
'come'. The verbs also have the correct termination markers. I
think the discrepancy in the types of causative
overgeneralizations produced in my spontaneous and experimental
studies might stem from the ages of the subjects in the two
studies. I plan to work with younger subjects in a future
experiment on the causative to see if they will use intransitive
verb forms in transitive sentence contexts.
(7) a. kamuxanik le: wich' aq
k-0-muxan-ik
INCOMP-3A-swim-IV the little pig
'the little pig is swimming.'
b. kuyo'oj b'ik eh kapetik
k-0-u-yo'-oj b'i-k eh k-0-pet-ik
INCOMP-3A-3E-scold-TV there-T and INCOMP-3A-come-IV
'She scolds him and then he comes.'
Theoretical Implications
I think this data, admittedly preliminary, supports a number
of conclusions about the process of acquiring the causative
alternation. Both the spontaneous and experimental data confirm
observations from English that children will overgeneralize the
causative alternation in ways that violate adult usage. This
finding disconfirms Baker's (1979) hypothesis of conservative
learning. Unfortunately this finding also raises the
learnability paradox discussed so extensively in Pinker (1989).
What mechanism can children rely upon to learn the adult
restrictions on lexical alternations like the causative?
Braine et al. (1990) propose a competition between the
verbs' argument structure and canonical sentence schemas to
account for children's causative overgeneralizations. This is
essentially identical to Pinker's hypothesis that children first
construct a broad-based rule as the basis of the causative
alternation. It is important to recognize that the alternation
takes place in both directions as Braine et al. demonstrate (see
also Lord 1979, Berman 1982.180 and Aksu-Koç & Slobin 1985.848).
Children could apply such a rule without changing the lexical
entries of verbs. However, some additional mechanism is needed
to explain why adults do not use canonical sentence schemas as
often as children. The fact that the K'iche' children in the
experimental study never used an intransitive verb in a
transitive argument frame suggests that they have extracted more
than a simple alternation between argument structures. They have
mastered the morphological changes associated with the changes in
verb transitivity in K'iche' (cf. Pye 1985). Pinker (1989)
proposes that children come to rely upon narrow range semantic
verb classes to acquire the adult restrictions on lexical
alternations like the causative. Pinker derives the narrow
semantic verb classes on the basis of his Grammatically Relevant
Subsystem Hypothesis (1989). He states that 'for it to be true,
there would have to be a single set of elements that is at once
conceptually interpretable, much smaller than the set of possible
verbs, used across all languages, used by children to formulate
and generalize verb meanings, used in specifically grammatical
ways ..., and used to differentiate the narrow classes that are
subject to different sets of lexical rules' (169). He fails to
show that the set of elements he uses to characterize the
causative alternation in English has crosslinguistic validity or
guides children in restricting the causative alternation to
specific narrow verb classes.
The form of Pinker's hypothesis creates a dilemma. If the
set of semantically relevant elements is too small, the degree of
crosslinguistic variability is overly restricted. It follows
that all languages would differentiate pretty much the same
narrow verb classes. If the set of elements is too large it will
allow for more crosslinguistic variation, but at the cost of
becoming nothing more than an extremely complex feature notation.
Unfortunately, it is difficult to tell how Pinker intends to use
his semantically relevant elements because many of his semantic
representations contradict his definitions of the semantic
elements. He represents the verb support as a STATE
incorporating an ACT (201). Earlier he states that ACTs have the
feature <+dynamic> while STATEs are <-dynamic> (195). Feature
clashes of this sort usually lead to ill-formed structures.
Another example of his flexibility in applying the basic
semantic elements occurs in his discussion of the crosslinguistic
differences that exist with respect to the
unergative/unaccusative distinction. By and large unaccusative
verbs causativize while unergative verbs do not. There is
considerable crosslinguistic variation in the unaccusative
distinction, however (Rosen 1984). Pinker solves this problem by
hypothesizing that in some languages ambiguous verbs 'may be
expressed as a kind of ACT, in others as a kind of GO or BE'
(225). Such flexibility indicates that Pinker's elements do not
have an independent conceptual interpretation; they can be
attached to any verb as a diacritic of its ability to
causativize. This imprecision makes it impossible to empirically
test Pinker's theory in its current form.
Apart from these definitional problems Pinker's theory
cannot account for the findings from the experimental study. His
theory does not predict crosslinguistic differences in which
verbs children will be prone to overgeneralize. All children
learning a particular verb should have the same difficulty in
learning the semantic elements associated with the verb's
meaning. Pinker's narrow range semantic constraints hypothesis
cannot explain why K'iche' children show such reluctance to
causativize the verb petik 'come' while English-speaking children
causativize the verb 'come' with relative abandon. Since the
meaning of these verbs is as similar as verb meanings can be in
different languages, some factor beyond verb meaning must affect
children's tendency to causativize verbs.
Another difficulty with Pinker's theory is that it is not
compatible with the extended period of time children require to
develop causative restrictions. Thirteen-year-olds should have a
fairly accurate representation of the meaning of muxanik 'swim'
and sutinik 'turn' as well as a basic understanding of which
narrow range semantic classes these verbs belong to. Although
Pinker is not specific about the length of time children need to
establish the narrow range restrictions on lexical rules such a
process should not require the better part of a decade. The
semantic elements that are basic to Pinker's acquisition
mechanism should be readily accessible to children since they
constitute a small set of universally relevant semantic features.
The preceding arguments suggest that lexical retrieval
processes may be the primary determinant of children's causative
errors. I think it is best to view the problem in the general
perspective of choosing a suitable verb on any given occasion.
Children are learning the difference between the verbs come and
go, bring and take. The causative alternation requires that they
also appreciate the difference between the verbs come and bring
as well as qajik and qasa:j. Several studies have suggested that
children do not always succeed in retrieving the proper word
(Hoek, Ingram & Gibson 1986). The retrieval process is
especially indicated in children's failure to select a suppletive
alternate. They may lack this alternate or not be able to
retrieve it as readily as the other form. The availability of
suppletive forms provides children with positive evidence for the
appropriateness of the different verbs. Thus acquiring lexical
alternations is no different from acquiring irregular
inflections. Pinker states that such suppletive alternations
account for 77% of Bowerman's data.
Another finding in favor of the retrieval process is that it
explains why children only produce causative errors
intermittently. If children actually did have an immature
semantic representation of a verb's meaning, they would use the
verb incorrectly every time they met a suitable occasion.
Instead, children only produce causative overgeneralizations in
extraordinary circumstances (such as elicitation experiments),
and then only in a certain percentage of instances. The
retrieval explanation would account for the individual
differences between children in the verbs they overgeneralize as
well as the frequency with which they overgeneralize them. Each
child's history of encounters with verbs would lead to individual
developmental profiles. The retrieval process would also become
better with time as children added suppletive forms to their
lexicon and strengthened their access to individual verbs. This
would be compatible with the extended developmental time frame
seen in the data.
I have mentioned before that another factor seems to be
affecting the children's access to particular verbs. The best
example of this is the difference between the K'iche' and
English-speaking children's willingness to causativize the verb
for 'come' in their languages. K'iche' children would be able to
use the monosyllabic form of the verb stem as additional
information about the verb and its possibilities for
participating in a transitivity alternation. There is a basic
split in the K'iche' language between monosyllabic and
polysyllabic verb stems. Monosyllabic stems are underived
transitive or intransitive stems. They only alternate with the
addition of an affix. Most polysyllabic verb stems are derived
from some other type of root. They are more likely to alternate
in transitivity with a simple affix change. The K'iche' children
could use the monosyllabic status of the verb pet to infer that
it was an underived intransitive verb and only alternate it when
they encountered positive evidence in their input.
The K'iche' children show further evidence of this
sensitivity to the derived/underived verb distinction in their
willingness to overgeneralize the causative alternation to the
verbs muxanik 'swim' and wakatik 'stroll'. Their tendency to
overgeneralize the verb muxanik is especially pronounced and may
stem from the misinterpretation of the /n/ in the stem as an
absolutive antipassive affix. Verbs with the antipassive have a
straight forward transitive form, and if muxanik was an
antipassive form, its transitive equivalent would be muxa:j.
This is indeed the form supplied most frequently by the K'iche'
children, and thus striking evidence that the children have
extracted the underlying distinction between derived and
underived verb stems.
I think there is support for this position in the results
Berman (1982) reports on verb-pattern alternation in Hebrew.
Recall that Hebrew children use intransitive verb forms in
transitive contexts in the same way English-speaking children do.
They also make the reverse error as Lord (1979) reports for
English. Hebrew employs a complex system of stem changes to
encode the causative alternation. Verb stems generally fall into
a set of five patterns that constitute the Binyan system of
alternations. Most of the verbs in pattern one (e.g. katav
'write') have a causative form in pattern five (e.g. hixtiv
'dictate'). Some pattern one verbs (e.g. lamad 'learn') have
causative forms in pattern three (e.g. limed 'teach'). Reflexive
verbs (e.g. mitracex 'wash') in pattern four have a transitive
form in pattern one (e.g. roxec) and inchoative verbs retain
their pattern five form in causative contexts (e.g. yavri 'become
healthy'). A few process verbs also have the same pattern five
form in intransitive and causative contexts (e.g. hitxil
'start').
This is a complex system and not surprisingly it takes many
years before children learn it completely. I think the
interesting feature of the Hebrew Binyan system is the extent to
which the first pattern contains many transitive and intransitive
verbs. Berman states that pattern one is used for the transitive
actions make = do, give, eat, drink, build, for statives such as
like, want, see, and for intransitives like go, sleep, run, jump
and sit. Thus, unlike the situation in K'iche' where there is a
clear morphological distinction between transitive and
intransitive verbs, children learning Hebrew have plenty of
evidence that pattern one contains both transitive and
intransitive verbs. I think this is the primary reason why they
use the verbs in different contexts without making the necessary
morphological changes.
My last conclusion would be that children may never entirely
succeed in accessing the correct verbs all of the time. I have
received written responses in my university classes from
undergraduates who have overgeneralized verbs. One such example
is 'These changes don't deteriorate the language'. Even the 1991
Stanford child language conference abstracts contain the example,
'When encountered with sentences ....' Thus the retrieval
process becomes essentially error free in adults for core verbs,
but remains susceptible to intrusions in the case of low
frequency verbs.
Acknowledgments
I presented preliminary versions of this paper at the 1990 Mid-
America Linguistics Conference, November 7, 1990, the 1991
Stanford Child Language Research Forum, April 7, 1991, and the
workshop on Crosslinguistic and Crosspopulation Contributions to
Theories of Language Acquisition, June 6, 1991, Jerusalem. I
thank all of these audiences for their suggestions, especially
Ruth Berman, Paul Bloom, Charles Ferguson, Jess Gropen, Teun
Hoekstra, David Ingram, and Dan Slobin. In addition Jess Gropen
responded to a preliminary version of this paper with an
extensive set of comments that were very helpful. None of these
people are responsible for the way I have incorporated their
suggestions in this version. Once again I must also acknowledge
the support my research receives in the Zunil community which
always makes fieldwork in Guatemala highly productive and
extremely enjoyable. I would hope that Zunil has recovered from
the devastating explosion that occurred there in 1991 as a result
of geothermal exploration.
Pedro Quixtan Poz has taught me much about when it is possible to
causativize K'iche' verbs and was instrumental in refining our
experimental techniques. My fieldwork in Guatemala was supported
by NSF grant No. BNS-8909846.
Notes
1All K'iche' words are shown in the practical orthography
developed by the Proyecto Lingüístico Francisco Marroquín
(Kaufman 1976) with a single exception: I use <'> rather than <7>
for the glottal stop. The other orthographic symbols have their
standard IPA values except: <tz> = /ts/, <ch> = /t�/, <b'> = /b/,
<tz'> = /ts'/, <ch'> = /t�'/, <x> = /�/, <j> = /x/. I use the
colon <:> to indicate long vowels.
I have also used the following morphological abbreviations:
COMP = completive aspect, INCOMP = incompletive aspect, 1A, 2A,
3A = first, second, third person singular absolutive person
markers (what Mayanists refer to as 'set B'), 1E, 2E, 3E = first,
second, third person singular ergative person markers (or 'set
A'), CAUSE = the causative morpheme, ABS = the absolutive
antipassive, TV = the affix marking derived transitive verbs, TTV
= the clause-final termination marker for root transitive verbs,
IV = the clause-final termination marker for intransitive verbs.
2This verb serves as an overt marker of the progressive aspect in
K'iche'. K'iche' speakers frequently shorten the verb to taq'en
or even q'en.
3Mayan linguists analyze the transitive verb form as containing
the causative suffix in an underlying level of the derivation.
Since Mayan verb roots are monosyllabic, all derivational
processes that yield transitive verb stems produce polysyllabic
stems. In this case the underlying form of the transitive verb
would be qa-isa-j. The initial vowel of the causative affix /i/
is lost through a regular process of vowel merger while the final
consonant /j/ of the intransitive stem would be added by
epenthesis. I have included the verb in the zero derivation
category since the processes of vowel merger and epenthesis
obscure the relation between the intransitive verb form and the
output of the affixal causative derivation. I also do not want
to assume that this relation is apparent to children learning
K'iche'.
4The translation of such verbs presents immense difficulties for
linguistic theories that seek a uniform mapping relation between
semantic and syntactic components. The K'iche' verb tze' might
also be translated as 'amuse', but the choice between 'laugh' or
'amuse' would have to be made on a semantic basis. To use verb
transitivity as a basis for translation would lead to a circular
argument that verb meaning determines verb argument structure and
verb argument structure determines the translation of verb
meaning (Pye, ms.).
5I will not examine antipassive forms of causativized verbs or
the causative forms of antipassive verbs in this paper. Both
occur sporadically in K'iche'.
6I used the first sample of Adam's speech (Adam01.chi) from the
Childes data base for this estimate (Brown 1973).
References
Aksu-Koç, A. A. & Slobin, D. I. (1985). The acquisition
of Turkish. In D. I. Slobin (ed.) The Crosslinguistic
Study of Language Acquisition. Vol. 1: The Data (pp. 839-878).
Hillsdale, NJ: Erlbaum.
Baker, C. L. (1979). Syntactic theory and the projection
problem. Linguistic Inquiry, 10, 533-581.
Baker, M. (1988). Incorporation: A theory of grammatical
function changing. Chicago: University of Chicago Press.
Berman, R. A. (1982). Verb-pattern alternation: The interface
of morphology, syntax, and semantics in Hebrew child language.
Journal of Child Language, 9, 169-191.
Bowerman, M. (1974). Learning the structure of causative
verbs: A study in the relationship of cognitive, semantic and
syntactic development. Papers and Reports on Child Language
Development, 8, 142-178.
Bowerman, M. (1982). Evaluating competing linguistic models
with language acquisition data: implications of developmental
errors with causative verbs. Quaderni di Semantica, 3, 5-66.
Braine, M. D. S., Brody, R. E., Fisch, S. M. & Weisberger, M. J.
(1990). Can children use a verb without exposure to its
argument structure? Journal of Child Language, 17, 313-342.
Brown, R. (1973). A first language. Cambridge, MA: Harvard
University Press.
Cazden, C. B. (1968). The acquisition of noun and verb
inflections. Child Development, 39, 433-448.
Chomsky, N. (1981). Lectures in the theory of government and
binding. Dordrecht: Foris.
Dayley, J. P. (1985). Tzutujil Grammar. Berkeley, CA: University
of California Press.
Green, G. M. (1974). Semantics and syntactic regularity.
Bloomington, Indiana: Indiana University Press.
Grimshaw, J. (1990). Argument structure. Cambridge, MA.: MIT
Press.
Hale, K. & Keyser, J. (1986). Some transitivity alternations in
English. (Lexicon Project Working Papers 7.) Cambridge, MA: MIT
Center for Cognitive Science.
Hoek, D., Ingram, D. & Gibson, D. (1986). An examination of
the possible causes of children's early word extensions. Journal
of Child Language, 13, 477-494.
Kaufman, T. 1976. Proyecto de alfabetos y ortografías para
escribir las lenguas mayances. Guatemala: Ministerio de
Educación.
Levin, B. & Rappoport Hovav, M. (1992). The lexical semantics
of verbs of motion: The perspective from unaccusativity. In
I. M. Roca (Ed.), Thematic structure: Its role in grammar (pp.
247-269). Berlin: Mouton de Gruyter.
Lord, C. (1979). "Don't you fall me down": Children's
generalizations regarding cause and transitivity. Papers and
Reports on Child Language Development, 17, 81-89.
Maratsos, M. P. (1979). How to get from words to sentences. In
D. Aaronson and R. Reiber (Eds), Psycholinguistic research:
Implications and applications. Hillsdale, NJ: Erlbaum.
Maratsos, M. P., Gudeman, R., Gerard-Ngo, P. & DeHart,
G. (1987). A study in novel word learning: the
productivity of the causative. In B. MacWhinney (Ed.),
Mechanisms of language acquisition (pp. 89-113). Hillsdale, NJ:
Erlbaum.
Mondloch, J. L. (1981). Voice in Quiché-Maya. Albany: State
University of New York. Unpublished Ph.D. dissertation.
Perlmutter, D. (1978). Impersonal passives and the
unaccusative hypothesis. In Proceedings of the fourth annual
meeting of the Berkeley Linguistics Society (pp. 157-189).
Berkeley: University of California.
Perlmutter, D. (1989). Multiattachment and the unaccusative
hypothesis: The perfect auxiliary in Italian. Probus, 1, 63-
119.
Perlmutter, D. & Postal, P. (1984). The one-advancement
exclusiveness law. In D. Perlmutter & C. G. Rosen
(Eds.), Studies in relational grammar, vol. 2 (pp.
81-125). Chicago: University of Chicago Press.
Pinker, S. (1989). Learnability and cognition: The
acquisition of argument structure. Cambridge, MA: MIT Press.
Pye, C. (1983). Mayan telegraphese: intonational determinants of
inflectional development in Quiché Mayan. Language, 59, 583-
604.
Pye, C. (1985). The acquisition of transitivity in Quiché
Mayan. Papers and Reports in Child Language Development, 24,
115-122.
Pye, C. (1987). The Pye Analysis of Language. Working Papers in
Language Development, 3, 1-37. The Child Language Program,
University of Kansas.
Pye, C. (1992). The acquisition of K'iche' Maya. In D.
I. Slobin (Ed.), The crosslinguistic study of language
acquisition, Vol. 3 (pp. 221-308). Hillsdale, NJ: Erlbaum.
Pye, C. ms. Language acquisition as 'radical translation'.
University of Kansas.
Pye, C. & Quixtan Poz, P. (1988). Precocious passives (and
antipassives) in Quiché Mayan. Papers and Reports on Child
Language Development, 27, 71-80.
Rosen, C. 1984. The interface between semantic roles and
initial grammatical relations. In D. Perlmutter & C. Rosen
(Eds.), Studies in relational grammar (pp. 38-77). Chicago:
University of Chicago Press.
Van Valin, R. D., Jr. (1990). Semantic parameters of split
intransitivity. Language, 66, 221-260.